Counterfactuals with feature preferences for consistent and diverse explanations

Abstract

An embodiment includes detecting by an artificial intelligence system a feature of a machine learning model. The embodiment includes responsive to detecting the feature of the machine learning model, computing by a Counterfactual Engine of the artificial intelligence system a counterfactual objective based on modifying a product of a weight and a perturbation of the feature. The embodiment also includes transforming by the Counterfactual Engine a counterfactual based on the counterfactual objective.

What is claimed is: 
     
         1 . A computer-implemented method comprising:
 detecting by an artificial intelligence system a feature of a machine learning model;   responsive to detecting the feature of the machine learning model, computing by a Counterfactual Engine of the artificial intelligence system a counterfactual objective based on modifying a product of a weight and a perturbation of the feature; and   transforming by the Counterfactual Engine a counterfactual based on the counterfactual objective.   
     
     
         2 . The computer-implemented method of  claim 1 , wherein the weight penalizes the perturbation of the feature. 
     
     
         3 . The computer-implemented method of  claim 1 , wherein the weight is a maximum value for the feature that is invariant. 
     
     
         4 . The computer-implemented method of  claim 1 , wherein the modifying further comprises minimizing the product based on a distance algorithm. 
     
     
         5 . The computer-implemented method of  claim 1 , wherein the weight comprises a cost preference. 
     
     
         6 . The computer-implemented method of  claim 1 , further comprising iterating the computing and the transforming wherein the iterating updates a set of counterfactuals and a current weighting vector. 
     
     
         7 . The computer-implemented method of  claim 6 , wherein the current weighting vector is computed based on a previous counterfactual from the set of counterfactuals. 
     
     
         8 . A computer program product comprising one or more computer readable storage media, and program instructions collectively stored on the one or more computer readable storage media, the program instructions executable by a processor to cause the processor to perform operations comprising:
 detecting by an artificial intelligence system a feature of a machine learning model;   responsive to detecting the feature of the machine learning model, computing by a Counterfactual Engine of the artificial intelligence system a counterfactual objective based on modifying a product of a weight and a perturbation of the feature; and   transforming by the Counterfactual Engine a counterfactual based on the counterfactual objective.   
     
     
         9 . The computer program product of  claim 8 , wherein the weight penalizes the perturbation of the feature. 
     
     
         10 . The computer program product of  claim 8 , wherein the weight is a maximum value for the feature that is invariant. 
     
     
         11 . The computer program product of  claim 8 , wherein the modifying further comprises minimizing the product based on a distance algorithm. 
     
     
         12 . The computer program product of  claim 8 , wherein the weight comprises a cost preference. 
     
     
         13 . The computer program product of  claim 8 , further comprising iterating the computing and the transforming wherein the iterating updates a set of counterfactuals and a current weighting vector. 
     
     
         14 . The computer program product of  claim 13 , wherein the current weighting vector is computed based on a previous counterfactual from the set of counterfactuals. 
     
     
         15 . A computer system comprising a processor and one or more computer readable storage media, and program instructions collectively stored on the one or more computer readable storage media, the program instructions executable by the processor to cause the processor to perform operations comprising:
 detecting by an artificial intelligence system a feature of a machine learning model;   responsive to detecting the feature of the machine learning model, computing by a Counterfactual Engine of the artificial intelligence system a counterfactual objective based on modifying a product of a weight and a perturbation of the feature; and   transforming by the Counterfactual Engine a counterfactual based on the counterfactual objective.   
     
     
         16 . The computer system of  claim 15 , wherein the weight penalizes the perturbation of the feature. 
     
     
         17 . The computer system of  claim 15 , wherein the modifying further comprises minimizing the product based on a distance algorithm. 
     
     
         18 . The computer system of  claim 15 , wherein the weight comprises a cost preference. 
     
     
         19 . The computer system of  claim 15 , further comprising iterating the computing and the transforming wherein the iterating updates a set of counterfactuals and a current weighting vector. 
     
     
         20 . The computer system of  claim 19 , wherein the current weighting vector is computed based on a previous counterfactual from the set of counterfactuals.